GitHub will rebuild its Git repository architecture from scratch to respond to a surge in traffic brought by AI agents, The Register reported on Thursday.

The report said the new architecture made write performance 35 times faster in internal tests. GitHub plans to keep developers' working methods, code reviews and security settings unchanged, and alter only the internal structure. It did not disclose a transition timeline.

As customers hand coding to AI agents, traffic to GitHub has also surged. Monthly events rose more than two-fold to 473.3 billion from 218.2 billion between September 2025 and August 2026. Commits in September totaled 7.38 billion, five times more than a year earlier. Service disruptions also occurred in succession during this period. In April alone, there were 10 performance degradation incidents, followed by 9 in May.

The existing repository architecture, called Spokes, stores full copies of a repository across multiple disks. Each time it stores new data, the job is completed only after multiple copies confirm they have finished storing it. It improves reliability, but it slows performance because it must wait for the slowest copy to confirm.

The new architecture writes data only once to Azure Blob Storage. Replication and backups are handled automatically by Azure Blob Storage. Separate lightweight servers handle read requests, while management tasks such as data compression and cleanup of unnecessary data run separately in the background.